The Accuracy of Activity Monitors Worn by Dairy Cows Housed in a Pasture-based Robotic Milking System

Behavioral observations of cattle are important for detection of illness, injury, and estrus as well as to verify that the cattle are behaving normally. However, conducting live observations of animals managed in extensive systems, such as pasture-based dairies, can be difficult and time consuming. Activity monitors, such as those developed for use with automatic milking systems (AMS), have been developed to automatically and remotely collect individual behavioral data. For example, in the Lely A3 Astronaut AMS, each cow wears a collar transponder (ACT) for identification, which can also collect data on her individual activity and rumination. The aim of this study was to examine whether cow activity levels reported by ACT are accurate compared to live observations (OB) and to levels reported by previously validated livestock pedometers (IceQube (IQ)). Fifteen lactating Holstein-cross cows with pasture access were fitted with ACT and IQ. Continuous focal observations were conducted (0600-2000) to describe the lying, standing, and walking behavior of the cows. The activity function in the IQ pedometers recorded the amount of time spent standing (S) lying (L), while IQ steps (ST) measured cow movement (i.e., acceleration of the cow as she moves around her environment). The ACT monitor reports the ‘activity,’ but it is unclear whether this measure of activity is measuring acceleration as a pedometer does or whether it includes standing behavior as well. Active behaviors (i.e., walking and standing) were analyzed separately and in combination (as activity) to determine exactly which behaviors contribute to calculation of the ACT monitor’s ‘activity’. Pearson correlations were performed between ACT and OB of walking (rP = 0.61, P < 0.0001), standing (rP = 0.47, P < 0.0001), lying (rP = -0.57, P < 0.0001), and activity (rP = 0.52, P < 0.0001). Pearson correlations for each cow were also performed between ACT and IQ steps (rP = 0.75, P < 0.0001), and IQ activity (rP = 0.58, P < 0.0001). Correlations were then analyzed with a general linear model to determine which comparisons were of similar strength. The ACT to OBW correlation was similar to both ACT to IQST and ACT to IQA (P > 0.05). The correlation between ACT and OBA was of similar strength compared with the correlation between ACT and OBS (P > 0.05). However, the correlation between ACT and IQST and the correlation between ACT and IQA were significantly different (P < 0.05). The correlation between ACT and OBL was significantly different (P < 0.0001) from all other correlations. Correlations between ACT and OBW, IQST or IQA were all significantly different (P < 0.05) than correlations between ACT and OBA or OBW. These data suggest ACT fails to distinguish between standing and lying behaviors, but reasonably represents cow walking.